부산대학교 도서관

Public buildings have huge energy consumption but incalculable energy saving potential, and the application of monitoring their load characteristics to explore energy saving potential is promising. Non-intrusive load monitoring and decomposition, as an advanced application on the distribution side of smart grids in the big data environment, can mine users' electricity consumption behavior through power port information, but traditional algorithms mainly target home users and have problems such as poor power traceability and long training time. For this reason, the paper proposes a load decomposition model based on deep learning and migration learning for building users. Based on this load decomposition technique, a proportional model based on the load decomposition technique is proposed for estimating the total power consumption increment of public buildings after electrical energy replacement. Firstly, we analyze the energy consumption scenarios of public building users and propose the appliances that can be used for electrical energy replacement. A load decomposition model for such appliances is built based on the public data set to construct the proportional relationship between the actual electricity consumption of non-all-electricity users and the incremental amount of electricity replaced. Then, based on the proportional relationship and the actual electricity consumption of the majority of known users in public buildings, the total replacement electricity increment after electric energy substitution is estimated. Finally, the estimated effect was measured using actual natural gas usage. The results show that it is inexpensive to sample only a small amount of electricity data from public buildings, and the obtained proportional relationship can better estimate the regional electricity consumption increment caused by electricity substitution, which has good application value in the implementation of electricity substitution.